Investigating possibilities of predictive mathematical models to identify at-risk students in the South African higher education context

Abstract

This article reports on the results of an investigation of the predictive accuracy of five different mathematical models to identify at-risk students in a Business Statistics course. Low levels of students’ success, especially in mathematics-related subjects such as statistics, are a salient problem in South Africa and other countries. Statistical knowledge is included in a variety of programmes offered by many faculties at tertiary level, and early prediction of at-risk students seems necessary to enhance academic success especially when dealing with large class groups. In this study, we used 395 Business Statistics students’ grades from an academic semester at an urban university in South Africa to build a predictive model to identify at-risk students. Grounded on Meyer’s model evaluation criteria and striving for a balance between accuracy and simplicity, two out of five models are identified as viable predictive models in identifying at-risk students by using a cross- validation test. The article shows the possibilities and limits in deriving information from a number of covariates. These results are interesting and have implications for educational practice in statistics courses.